From 19ccf930858e4db9b18479532450383e40d56897 Mon Sep 17 00:00:00 2001
From: Pinghao Wu <xdavidwuph@gmail.com>
Date: Mon, 20 Jan 2025 21:02:55 +0800
Subject: [PATCH] grafana: kubernetes-containers: handle crashloop better

---
 monitoring/grafana/kubernetes-containers.yaml | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/monitoring/grafana/kubernetes-containers.yaml b/monitoring/grafana/kubernetes-containers.yaml
index c9455bb..1bd5ad4 100644
--- a/monitoring/grafana/kubernetes-containers.yaml
+++ b/monitoring/grafana/kubernetes-containers.yaml
@@ -27,7 +27,7 @@ spec:
           datasource='$DS_PROMETHEUS',
         ).addTarget(
           grafana.prometheus.target(
-            'label_replace(rate(container_cpu_usage_seconds_total{' + matchers + '}[$__rate_interval]), "host", "$1", "instance", "([^.]*).*:10250")',
+            'label_replace(sum by (namespace, pod, container, instance) (rate(container_cpu_usage_seconds_total{' + matchers + '}[$__rate_interval])), "host", "$1", "instance", "([^.]*).*:10250")',
             legendFormat='{{pod}}/{{container}}@{{host}}',
           )
         ),
@@ -40,7 +40,7 @@ spec:
           format='bytes',
         ).addTarget(
           grafana.prometheus.target(
-            'container_memory_usage_bytes{' + matchers + '}',
+            'sum by (namespace, pod, container) (container_memory_usage_bytes{' + matchers + '})',
             legendFormat='{{pod}}/{{container}}',
           )
         ),
@@ -53,7 +53,7 @@ spec:
           format='bytes',
         ).addTarget(
           grafana.prometheus.target(
-            'container_memory_rss{' + matchers + '}',
+            'sum by (namespace, pod, container) (container_memory_rss{' + matchers + '})',
             legendFormat='{{pod}}/{{container}}',
           )
         ),
@@ -66,7 +66,7 @@ spec:
           format='bytes',
         ).addTarget(
           grafana.prometheus.target(
-            'container_memory_cache{' + matchers + '}',
+            'sum by (namespace, pod, container) (container_memory_cache{' + matchers + '})',
             legendFormat='{{pod}}/{{container}}',
           )
         ),
@@ -79,7 +79,7 @@ spec:
           format='Bps',
         ).addTarget(
           grafana.prometheus.target(
-            'rate(container_network_receive_bytes_total{interface="eth0", ' + podMatchers + '}[$__rate_interval])',
+            'sum by (namespace, pod, container) (rate(container_network_receive_bytes_total{interface="eth0", ' + podMatchers + '}[$__rate_interval]))',
             legendFormat='{{pod}}',
           )
         ),
@@ -92,7 +92,7 @@ spec:
           format='Bps',
         ).addTarget(
           grafana.prometheus.target(
-            'rate(container_network_transmit_bytes_total{interface="eth0", ' + podMatchers + '}[$__rate_interval])',
+            'sum by (namespace, pod, container) (rate(container_network_transmit_bytes_total{interface="eth0", ' + podMatchers + '}[$__rate_interval]))',
             legendFormat='{{pod}}',
           )
         ),
-- 
2.45.2